CS386_coursepro

The course project for CS386

vgg16 pretrain is needed.
Due to the file size limitation, it is not uploaded.
The relative files is uploaded via BaiduNetDisk.
Also, for RBD part, RBD prior is needed. The images should be prcessed by the matlab code uploaded.

Environment

We basically implement the codes in Pytorch4.0.

Usage

Type python3 main.py in the console in each sub directory.

Dataset

We use MSRA-B 5k as our training dataset. We also manually divde MSRA-B into two parts, a 4000 sub part is used for training and the other 1000 is used for validation. We also use ECSSD and HKU-IS dataset to validate the efficiency of our model.

Results

F-score on different datasets:

Tables	MSRA-B validation	ECSSD	HKU-IS
E-HED+RBD	0.9329	0.9268	0.9305
E-HED+BRN	0.9302	0.9244	0.9267

MAE on different datasets:

Tables	MSRA-B validation	ECSSD	HKU-IS
E-HED+RBD	0.0593	0.0414	0.0446
E-HED+BRN	0.0548	0.0408	0.0414